Using Wiktionary to Create Specialized Lexical Resources and Datasets

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3457–3460 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3457 Using Wiktionary to Create Specialized Lexical Resources and Datasets Lenka Bajˇceti´c1, Thierry Declerck2 1 Austrian Centre for Digital Humanities and Cultural Heritage, Austrian Academy of Sciences, Vienna, Austria 2 DFKI GmbH, Multilinguality and Language Technology Lab, Saarland University Campus D3 2, Germany 1Lenka.Bajcetic@oeaw.ac.at, 2declerck@dfki.de Abstract This paper describes an approach aiming at utilizing Wiktionary data for creating specialized lexical datasets which can be used for enriching other lexical (semantic) resources or for generating datasets that can be used for evaluating or improving NLP tasks, like Word Sense Disambiguation, Word-in-Context challenges, or Sense Linking across lexicons and dictionaries. We have focused on Wiktionary data about pronunciation information in English, and grammatical number and grammatical gender in German. Keywords: Wiktionary, ambiguities, pronunciation,grammatical number, grammatical gender 1. Introduction In this paper, we investigate the use of Wiktionary (Wikimedia, 2021b) for building (lexical) datasets that can support the improvement and/or evaluation of challenging NLP tasks, like Word Sense Disambigua- tion (WSD), Word-in-Context challenges (WiC), Sense Linking (SL), or Machine Translation (MT). Extracting lexical information from Wiktionary can also be used for enriching other lexical resources. Wiktionary is a freely available web-based multilin- gual dictionary for over 170 languages, supported by the Wikimedia Foundation (Wikimedia, 2021a). Wik- tionary provides detailed information on lexical entries, which may include inflectional and derivational infor- mation, definitions, examples of usage, pronunciation, etymology, translations, to name just the more relevant for our experiments. Like other Wikimedia (Wikimedia, 2021a) supported initiatives, Wiktionary is a collaborative project. This implies that there might be inaccuracies in the resource, but the editing system is helping in mitigating this risk. The fact that Wiktionary is built by a collabo- rative effort means that the coverage and variety of lex- ical information is much larger than any single curated resource, while Wiktionary is integrating information from expert-based dictionary resources, when their li- censing conditions allow it. In addition to its availability, Wiktionary is easily ac- cessible and well-structured. Since Wiktionary grows ‘organically’ as a product of collaborative contribu- tions, there are inevitably some variations in data orga- nization and not all language versions follow the same structure for encoding their data. Wiktionary provides a range of data-dumps in XML, which we access using Python scripts. Each lexical en- try has its own Wiktionary page and in the XML dump its own <page>-tag, individual words can be found within the <title>-tag, and parts containing the lex- ical information are available within the <text>-tag. But the information within the <text>-tag is encoded by using a wiki mark-up language. The structure of information at this last tag, however, is not the same throughout all languages, so that each language spe- cific XML dump file has to be accessed individually. We focus on linguistic properties of lexical entries that are often omitted from other resources, especially from lexical semantics resources, and can …

Content truncated to 500 words